Quantification flags (quant_flag)

Every fitted or quantified spectrum is assigned an integer quant_flag. The flag records whether AutoEMX considers the result reliable and, if not, why. Flags are stored on each spectrum’s quantification record (and appear as Quant_flag in legacy Data.csv outputs).

At analysis time, quant_flags_accepted (in the clustering configuration) decides which of those flags are kept for compositional clustering. Changing quant_flags_accepted does not re-run quantification; it only changes which already-quantified compositions enter the clustering step.

Default for analysis workflows is quant_flags_accepted = [0, -1].

Flag meanings

Flag

Meaning

0

Quantification is OK (may still carry large analytical error).

-1

Same as 0, but quantification did not converge within the iteration limit.

1

Error during EDS acquisition / no spectral data. No fit executed.

2

Total counts below 90% of the target acquisition counts (often bad segmentation or a truncated acquisition). Fitting is skipped when interrupt_fits_bad_spectra=True.

3

Too little low-energy signal (background under ~2 keV too low), which compromises P/B ratios. Fitting is skipped when interrupt_fits_bad_spectra=True.

4

Poor fit (reduced chi-squared too high relative to total counts). Fit may be interrupted when interrupt_fits_bad_spectra=True.

5

Excessively high analytical error (> 50 wt%). Often caused by a missing element or another major model error. Fit may be interrupted when interrupt_fits_bad_spectra=True.

6

Excessive X-ray absorption (particle geometry). Fit may be interrupted when interrupt_fits_bad_spectra=True.

7

Excessive substrate contamination (a substrate-element peak intensity above 10% of total counts).

8

Background counts under a reference peak are too low (i.e. below min_bckgrnd_cnts).

9

Fit interrupted for an unknown reason.

10

Experimental-standards measurements only: a required reference peak is missing.

How to use them

  • Prefer keeping [0, -1] for phase identification unless you intentionally want to inspect poorer spectra.

  • To include borderline spectra in clustering, add their flags to quant_flags_accepted (for example, you may consider including 8, which is often quite accurate anyways (see Fig. 4 in the paper)).

  • Prefit issues (1, 2, 3) can stop fitting early when interrupt_fits_bad_spectra=True. Flags 2 and 3 can still be written on a completed quantification if fitting was allowed to proceed.

Related parameters: interrupt_fits_bad_spectra, min_bckgrnd_cnts, and max_analytical_error_percent. See Tutorial: EDS compositional analysis for phase identification and Tutorial: Quantify externally exported spectra in batch.